The "AI" phenomenon and craze has been discussed endlessly, including the vast computational and energy requirements of dedicated #datacenters
There is an important split betwen model building and inference of statistical models (and hence also #llm ) which means there is not one but two problems to solve.
The first one is *highly* non-trivial. It bumps against #mooreslaw and requires #supercomputing networking tricks. The datacenter must operate as one giant machine
Using large (or largish) models locally (so called #localAI ) falls under the second, inference task. It is an emerging passtime for technical people with deeper pockets (one still needs a fairly massive server). It may become more of a commodity thing in due course .
Democratized building of large models without access to ultra-expensive and complicated supercomputing is theoretically possible. Ceteris-paribus It will be slower and it needs some breakthroughs in large scale #federatedlearning
Open Risk